Understanding an irregular pattern of fracture slip from laboratory earthquake using machine learning

Abstract

A series of direct-shear experiments on a sawcut fracture under a multi-stage reduction in normal stress shows slip transition behaviors from stick-slip to stable sliding. The irregular slip pattern near the stability transition is traditionally considered unexplainable and unpredictable. This study utilizes a support vector machine (SVM) to classify the fracture slip behaviors and to predict new, unseen slip events. The results exhibit that the SVM trained by the acoustic emission amplitude well predicts the variation of shear stress near the stability transition, and the prediction accuracy slightly decreases with more irregular slip events occurred in the transition stage. The prediction for a fast rupture is more accurate than that for a slow rupture as relatively weak signals from a slow rupture are more sensitive at the same attenuation level. Understanding the irregular slip pattern allows us to better explain the occurrence of Parkfield tremors in California and provides more insights into the mechanisms of repeating earthquakes.